@yonderbarn@lazysoci.al to Technology@lemmy.worldEnglish • 8 days agoSilicon Valley Is Panicking About Zohran Mamdani. NYC’s Tech Scene Is Notwww.wired.comexternal-linkmessage-square109fedilinkarrow-up1639arrow-down110cross-posted to: hackernews@lemmy.bestiver.se
arrow-up1629arrow-down1external-linkSilicon Valley Is Panicking About Zohran Mamdani. NYC’s Tech Scene Is Notwww.wired.com@yonderbarn@lazysoci.al to Technology@lemmy.worldEnglish • 8 days agomessage-square109fedilinkcross-posted to: hackernews@lemmy.bestiver.se
minus-square@MonkderVierte@lemmy.ziplinkfedilinkEnglish27•edit-27 days agoLook, that character switch trick doesn’t poison any AI* but it’s annyoing to read. * Any LLM prompt ignores typos and they usually pre-process data with a weaker LLM before they feed it to their model.
minus-square@Bassman1805@lemmy.worldlinkfedilinkEnglish8•7 days agoIt’s not an LLM poisoning thing, they just legitimately believe in bringing back the thorn character. I agree, it’s ineffective and annoying.
minus-square@mcv@lemmy.ziplinkfedilinkEnglish3•edit-26 days agoI like the idea, but I’m annoyed by the inconsistency. The þ is the th sound in “both”, but not the th sound in “the”; that’s a ð. Ðough, ðat, ðere Þorough, boþ, þree
Look, that character switch trick doesn’t poison any AI* but it’s annyoing to read.
* Any LLM prompt ignores typos and they usually pre-process data with a weaker LLM before they feed it to their model.
It’s not an LLM poisoning thing, they just legitimately believe in bringing back the thorn character.
I agree, it’s ineffective and annoying.
I like the idea, but I’m annoyed by the inconsistency.
The þ is the th sound in “both”, but not the th sound in “the”; that’s a ð.
Ðough, ðat, ðere
Þorough, boþ, þree